Hu, Dong (The Hong Kong Polytechnic University), Huang, Chao (The Hong Kong Polytechnic University), Huang, Hailong (The Hong Kong Polytechnic University)

Enhancing Data-Driven Energy Management Strategy Via Digital Expert Guidance for Electrified Vehicles

Scheduled for presentation during the Regular session "Energy Management" (FrB2), Friday, November 1, 2024, 11:30−11:50, Room T2

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on January 2, 2025

Abstract

This study addresses the data efficiency and reliability issues in the energy management strategy (EMS) based on reinforcement learning (RL) for hybrid electric vehicles (HEVs). It proposes a novel expert-guided RL (EGRL) paradigm, which combines deep ensemble methods and digital expert model (DEM) to achieve real-time intervention and demonstration of EMS in various scenarios, thereby effectively enhancing system performance. Specifically, DEM is trained through domain adversarially invariant meta-learning (DAIML) to adapt to different driving scenarios. An intervention mechanism is established based on uncertain evaluation in the deep ensemble scheme, enabling DEM to supervise and intervene in the training process of RL, providing reliable guidance. Simulation results demonstrate significant effectiveness of the proposed paradigm in energy management tasks, approaching optimal results in various driving scenarios and outperforming traditional RL methods. The overall economic benefit of EGRL is approximately 15.8% better on average than traditional RL across all test cycles.